Summary
Veteran AI introduces LTX Director, a new node in the LTX 2.3 ComfyUI ecosystem developed by the WhatDreamsCost repository that consolidates multiple video generation control methods into a single timeline-based interface. Building on the earlier Prompt Relay concept — which maps different text prompts to different temporal segments — LTX Director adds support for starting reference images, ending reference images, multiple middle-frame anchors, per-segment prompts, and audio, all configured from one node. Installation is handled through ComfyUI Manager or by manually cloning into the custom_nodes folder, with KJNodes and ComfyUI-LTXVideo as required dependencies.
The video goes beyond a simple walkthrough by presenting an improved workflow that outperforms the official release. The two key changes are loading the main model via UNETLoader (which has distillation built in, eliminating the need for an acceleration LoRA) and replacing BasicScheduler in the upscaling stage with ManualSigmas while setting denoise to 1.0 instead of the official 0.42. Side-by-side comparisons show the improved version produces noticeably smoother motion with a larger range of movement and less rigidity.
Practical test cases covered include basic image-to-video, multi-segment Prompt Relay-style generation with a global character prompt plus local scene prompts, and switching to an uncensored model variant for more dynamic motion. Workflows are available to replicate directly on RunningHub, making this a useful reference for ComfyUI users already working within the LTX 2.3 pipeline.
📺 Source: Veteran AI · Published May 25, 2026
🏷️ Format: Tutorial Demo
